| The Biostatistics group is focused on the modeling of medical data, whether pre-clinical, clinical or medical instrumentation. Within REDM the contribution is especially aimed at drawing inference and offering interpretability based on data of both experiments and high dimensional data. For the research community we additionally offer support through consultancy at the core-facility SQUARE. We also take up our responsibility for the education of our students. |
Our team
ResearchersKurt Barbé, PhD. InternsNiels Cleymans | Teaching StaffRobin Van den Bossche Consultants | Former StaffSusanne Blotwijk, PhD. 09-2023 |
running projects
Mapping care pathways in children with TBI using machine learning
V. De Deken (K. Putman - K. Barbé)
In general, brain injuries present a significant societal burden. Focus should be brought not only to prevention but also towards improved quality of care. Currently, there is too much unexplainable variation in care trajectories. One category of patients less studied in this domain is pediatric traumatic brain injuries. To overcome this problem, administrative data holds the potential to provide a solution. In this project, we aim to identify different care pathways using machine learning techniques in order to better understand the variation in care and work towards improving the quality of care received by patients.
An artificial intelligent consultant to enhance statistical thinking
P. Savieri (K. Barbé - L. Stas)
Statistical consultants frequently encounter researchers seeking support for their study design and data analysis. Such requests for support often encompass doubts in terms of the type of analysis warranted. Although academic researchers obtained specific training in quantitative courses, the lack of hands-on experience often makes researchers feel uncomfortable with respect to statistical choices and their data analyses. In this project, we intend to develop an artificial intelligence (AI) system to provide feedback to the researcher. The tools will be interactive web applications built from the Shiny package within the statistical R environment. The apps will not be exhaustive, but will provide the necessary scientific tutoring, references and reflections for specific statistical techniques/assumptions which are considered healthy thinking approaches.
Radiomics in cancer radiotherapy: unraveling the biology to optimize treatment
A. Rifi (K. Barbé - M. De Ridder)
Radiomics can be used to provide valuable information for personalized cancer therapy. However, the dimensionality implies that radiomics renders a huge data set with many quantitative variables extracted from CT images of the tumor. One way to deal with this dimensionality is by applying random forests as a multivariate classification tool to analyze the radiomics data to predict tumor response. Although such random forests can cope with high dimensional data, its predictions are a black box system. In this project, we will break in into the black box. By dedicated animal trials, we will explore the decision-making process of the random forest to understand what triggers specific predictions.
Random neural network forests decision support for markerless stereotactic body radiotherapy
C. Raets (K. Barbé - M. De Ridder)
In the field of dosimetry, clinicians define the correct dose of radiotherapy to treat cancer patients. In this project, we examine and design the potential of random forests for the dose calculation. By acknowledging that neural networks allow modelling complex relationships; we wish to combine random forests and neural networks such that an ensemble of 'simpler' neural networks, yet with random weights, can be obtained to alleviate potential overfitting leading to more precise dose calculations.
finished projects
Adaptive design in preclinical research
S. Blotwijk & K. Barbé
Several forms of adaptive designs have been developed to optimize experimental designs, to minimize required sample sizes, to guarantee sufficient data, to most efficiently identify valuable research domains, and for many other reasons. Unfortunately, the statistical methodology is usually not suitable for the small sample sizes. We set out to remedy this situation by developing appropriate methodology and making it available through open source software, such that preclinical researchers can immediately apply it in practice.
PhD. Susanne Blotwijk, 2023
Fractional-order Wiener-Hammerstein systems for bio-impedance spectroscopy
H. Shaikh & K. Barbé
Wiener-Hammerstein models are a popular class of nonlinear dynamic time series as a simple type of nonlinear Volterra model. The Volterra model holds the same properties as what is observed in linear time series for nonlinear time series. As a result, the nonlinear system is decomposed into a purely linear and purely nonlinear part. In this project, we allow the linear dynamics to show fractional behaviour specifically of interest in the field of (bio)spectroscopy. Moreover, the project aims for algorithmic innovations which speeds up the training of the Wiener-Hammerstein model considerably.
PhD. Hanif Shaikh, 2022
Akritas-Arnold tests to assess nanoParticles to Activate immune Responses against Cancer
O. Olarte & K. Barbé
Contrast tests achieve a high power but unfortunately non-parametric contrast tests are limited to the Jonckheere-Terpstra test. Emerging tests of the Akritas-Arnold type is not fully understood and explored. This family of tests can be positioned within the rank randomization tests. In this project, we aim at a better understanding of the mathematical properties of the Akritas-Arnold tests which will be validated on a preclinical study on immunotherapy against cancer.
Postdoc project, finished 2023
